Characterization of a thermodenuder-particle beam mass spectrometer system for the study of organic aerosol volatility and composition

Characterization of a thermodenuderparticle beam mass spectrometer system for the study of organic aerosol volatility and composition A. E. Faulhaber, B. M. Thomas, J. L. Jimenez, J. T. Jayne, D. R. Worsnop, and P. J. Ziemann Air Pollution Research Center, University of California, Riverside, California, USA Department of Chemistry and Biochemistry, and Cooperative Institute for Research in the Environmental Sciences (CIRES), University of Colorado, Boulder, Colorado, USA Aerodyne Research Inc., Billerica, Massachusetts, USA Received: 4 August 2008 – Accepted: 19 August 2008 – Published: 4 September 2008 Correspondence to: P. J. Ziemann (paul.ziemann@ucr.edu) Published by Copernicus Publications on behalf of the European Geosciences Union.

and other quantitative aerosol mass spectrometers. The data that can be obtained are valuable for modeling organic gas-particle partitioning and for gaining improved composition information from aerosol mass spectra. The method is based on an empirically determined relationship between the thermodenuder temperature at which 50% of the organic aerosol mass evaporates (T 50 ) and the organic component vapor pressure at 10 25 • C (P 25 ). This approach avoids the need for complex modeling of aerosol evaporation, which normally requires detailed information on aerosol composition and physical properties. T 50 was measured for a variety of monodisperse, single-component organic aerosols with known P 25 values and the results used to create a log P 25 vs. T 50 calibration curve. Experiments and simulations were used to estimate the uncertainties in P 25

Introduction
The volatility of atmospheric organic aerosol (OA) has been the subject of considerable attention recently Robinson et al., 2007;Jonsson et al., 2007;Paulsen et al., 2006;Stanier et al., 2007;Huffman et al., 2008a 1 ). It not only affects the mass concentration and composition of OA subjected to changing environments 5 directly through gas-particle partitioning, but can also have a significant impact on aerosol chemistry. For example, it has been suggested (Robinson et al., 2007) that secondary organic aerosol (SOA) formed from the oxidation of semivolatile organic compounds that evaporate when primary organic aerosol (POA) is diluted in the atmosphere may explain recent field measurements of SOA concentrations well in excess 10 of those predicted by models (de Gouw et al., 2005;Heald et al., 2005;Johnson et al., 2006;Volkamer et al., 2006). The idea of incorporating realistic volatility behavior into OA models by sorting the OA mass into bins based on volatility (Donahue et al., 2006) has had some success in bringing modeled geographic distributions of organic aerosol into agreement with ob-15 servations (Robinson et al., 2007). In this scheme, components are binned according to their effective saturation concentrations, which can be estimated very simply from the vapor pressures of the pure components. A reasonably accurate description of the volatility behavior of the OA can be achieved by allowing each bin in the "volatility basis set" to cover one order of magnitude in effective saturation concentration. The 20 distribution of mass within (gas vs. particle) and among the bins changes with emissions, dilution, temperature, and chemical transformation, with the fraction of mass in each bin that is in the particle phase depending on the effective saturation concentration and the total OA mass concentration according to gas-particle partitioning theory measuring particle vapor pressure distributions is clear. A thermodenuder (TD), which is a flow-through system consisting of a heated vaporizer section in which particles evaporate, followed by a denuder section in which the vapor is removed by adsorption onto activated charcoal, is a useful tool for such measurements.
The Aerodyne Aerosol Mass Spectrometer (AMS) (Jayne et al., 2000;Jimenez et al., 2003) is widely used for mass spectrometric analysis of particulate matter in ambient studies. Its use in volatility studies to monitor changes in OA composition due to evaporation in a TD is practical, since the AMS can quantify total OA as well as specific OA components such as oxygenated OA (OOA) and hydrocarbon-like OA (HOA) (Zhang et al., 2005;Ulbrich et al., 2008). Two advantages of combining mass spec-15 trometric detection with volatility measurements are apparent. First, relationships can be determined between composition and volatility in the aerosol being studied, allowing greater insight into the chemistry and therefore origin and chemical evolution of different volatility fractions. Second, the mass spectrum is simplified by the separation of volatility-resolved fractions. Atmospheric aerosol is generally an extremely complex 20 mixture, and the composition of the organic fraction in particular is not well known or easy to characterize. A means of separating aerosol constituents online allows more information to be extracted from the mass spectra.
In this paper, we describe the characterization of a thermodenuder coupled to a thermal desorption particle beam mass spectrometer (TDPBMS) (Tobias et al., 2000), Introduction Conclusions References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion basis set analysis of the type used by Donahue et al. (2006) is used to show an alternative representation of the volatility distributions of these mixtures, and to predict their gas-particle partitioning. In addition, uncertainties in estimated vapor pressures, especially those due to the effects of OA mass concentration, particle size, and mixing state, which we have investigated through experiments and simulations, are discussed.

Aerosol generation
Monodisperse aerosol particles were generated by atomizing a 0.05 to 0.6 volume % 15 solution of the compounds of interest in 2-propanol. The solution was nebulized using a Collison atomizer with clean, dry air (RH<1%, total hydrocarbons <5 ppb) from an Aadco pure air generator. The resulting aerosol passed through two diffusion dryers filled with activated charcoal and a 210 Po bipolar charger before being size selected using a differential mobility analyzer (DMA). The number density was measured at 20 the beginning and end of each experiment using a Faraday cage aerosol electrometer positioned after the DMA. Polydisperse oleic acid aerosol particles were generated using an evaporation/condensation particle generator. Pure oleic acid was evaporated in a heated flask 25 Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion into a stream of nitrogen and then mixed with another stream of nitrogen to initiate particle formation by homogeneous nucleation. SOA was generated in a ∼6000 L PTFE environmental chamber. The chamber was initially filled with clean, dry air. For the reaction of pentadecane with OH radicals in the presence of NO x , 0.2 ppmv pentadecane, 10 ppmv methyl nitrite [CH 3 ONO], and 5 10 ppmv NO were added to the chamber and irradiated with blacklights to produce OH radicals (Atkinson et al., 1981). The blacklights were left on for 23 min to reach a mass concentration of ∼200 µg m −3 . The mass concentration was measured using an SMPS (Wang and Flagan, 1990) comprised of a long differential mobility analyzer, a 210 Po bipolar charger, a TSI Model 310 CPC, and scanning software provided by the 10 McMurry group at the University of Minnesota.

Thermodenuder
The TD design, depicted in Fig. 1, is similar to that described by Wehner et al. (2002) and is described in detail by Huffman et al. (2008b). It consists of a heated vaporizer section in which particles are volatilized, followed by a denuder section containing 15 activated charcoal to remove the vapors. Each section is about 50 cm long. The vaporizer is heated using three heaters, each of which is independently regulated using a PID controller to achieve a fairly uniform temperature profile. Temperature feedback to the PID controllers is provided by thermocouples measuring the temperature on the exterior surface of the heating tube. The controllers were set to produce equal wall tem-20 perature readings for all three heating zones, which required set-points slightly higher than the wall temperature. For example, temperature set-points of 152.6, 150.8 and 153.5 • C for the first, second, and third heating zones, respectively, were required for a wall temperature of 150 • C. The temperature profile within the vaporizer section of the TD was measured at several wall temperatures from 40 to 150 • C using a thermocouple 25 mounted in a 1/4 inch diameter stainless steel tube. The thermocouple was positioned in the flow and out of contact with the inner wall, at a series of measured locations along the length of the vaporizer. A flow rate of 0.6 l min −1 , the same as that used in the aerosol volatility experiments, was used for this characterization. The resulting centerline temperature profiles are shown in Fig. 2. The profiles show an initial temperature rise, followed by a small bump, then a plateau before the temperature falls at the end of the heated region. The temperature in the plateau is ∼1-2 • C below the wall temperature. For a wall temperature of 150 • C, at which the differences between the 5 wall and centerline temperature are the greatest, the highest temperature in the initial bump is ∼14 • C above the wall temperature, or ∼ 3% in terms of absolute temperature. These temperatures are somewhat lower and less uniform than those reported by Huffman et al. (2008b), who found centerline temperatures ∼17% above the setpoint measuring from room temperature for a TD of similar design (the TD used in this 10 study was a prototype, and that used by Huffman et al. was built using feedback based on this model). The absolute temperatures are within 5% of the wall temperature for a distance of ∼40 cm between the cooler end regions. Aerosol was sampled from either the atomizer/DMA or the environmental chamber, and, depending on the valve position, passed through either the TD or a bypass tube. A 15 portion of the aerosol stream was then directed into the TDPBMS. The flow rate through the TD system was 0.6 l min −1 , set by adjusting a valve located directly upstream of a diaphragm pump. The resulting effective plug flow residence time in the central 40 cm of the vaporizer section was ∼15 s at room temperature. The flow rate was regularly measured with a Sensidyne Gilibrator. M T , the aerosol mass concentration measured 20 at the exit of the TD when set at temperature T , and M 0 , the aerosol mass concentration measured at the exit of the TD bypass tube, were used to calculate the aerosol mass fractions remaining at a particular TD temperature, M T /M 0 . These values were the basis of the analysis employed in this study, and a TD vaporization profile consists of a plot of M T /M 0 vs. T . Both changes in signal intensity, which occur due to changes 25 in the aerosol mass concentration and signal drift in the mass spectrometer, and background signal must be accounted for in calculating M T /M 0 . The background signal was measured by setting the DMA voltage to 0 for monodisperse aerosols (so that no particles exit the DMA), or by placing a Teflon filter in the line upstream of the TD for Interactive Discussion polydisperse aerosol and SOA. It was measured frequently during the experiment, and the appropriate value to subtract from the signal at any time was estimated by interpolation. Background was subtracted from all signal intensities used in the calculations. In order to minimize the error due to drift in the aerosol signal over time, each pair of signal intensities used to calculate one value of M T /M 0 was measured within a period 5 of 4 to 5 min. At each TD temperature, the flow was directed through the TD for approximately 4 min. The signal measured at the beginning of the TD segment was divided by that measured just before the flow was switched from the bypass tube to the TD, and the signal measured at the end of the TD segment was divided by that measured just after the flow was switched back to the bypass tube (except for a period of about 10 90 s for the signal to equilibrate after switching each time). These two values were averaged to get a value of M T /M 0 for that temperature. Between TD segments, the flow was directed through the bypass tube for ∼6-10 min. Finally, M T /M 0 was corrected for the temperature-dependent particle losses in the TD, as described by Huffman et al. (2008b).

15
2.4 Thermal desorption particle beam mass spectrometer The TDPBMS used in this study has been described in detail previously (Tobias et al., 2000), and will only be described here briefly. The aerosol is sampled through a 0.1 mm critical orifice and passes through a series of aerodynamic lenses that focus the particles into a beam. The beam then passes through a nozzle and two flat-plate 20 skimmers and into the detection chamber, where particles impact on a V-shaped notch in a resistively heated copper vaporizer coated with a non-stick polymer. A fraction of the vaporized material diffuses into an ABB Extrel MEXM 500 quadrupole mass spectrometer and is ionized by 70 eV electrons, mass analyzed, and detected using a pulse-counting detector. In the experiments described here, the vaporizer was held at 25 a temperature of 160 • C in order to vaporize all organic aerosol components rapidly and obtain mass spectral data in real time. For the pure compounds used for calibration and the simple mixture, the signal intensity at a few strong peaks was monitored in 28 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion single ion monitoring (SIM) mode. For SOA, complete scans were recorded, and the TI (total ion) signal calculated for masses between m/z 45 and an upper limit between m/z 260 and 400, depending on the aerosol composition.
3 Results/analysis/discussion 3.1 Thermodenuder vaporization profiles 5 Figure 3 shows a plot of M T /M 0 , the fraction of the particle mass remaining after heating in the TD, vs. TD temperature for three dicarboxylic acids along with sigmoidal fits to the data. A plot of M T /M 0 vs. TD temperature will be referred to as a TD vaporization profile. The values of T TD on the x-axis refer to the temperatures measured on the outside of the TD flow tube, i.e., the wall temperatures. As mentioned above, the temperatures measured in the flow are within 15% of the wall temperatures for a distance of about 40 cm within the TD, with the remainder of the length of the TD heating region consisting of the temperature rise and fall regions. T 50 , the temperature at which half of the aerosol mass has evaporated, is a convenient temperature with which to characterize a pure standard compound. The tem-15 perature at the midpoint of the sigmoidal fit is used to determine T 50 for the standard compounds. While the TD vaporization profiles are not strictly sigmoidal, the fit allows for variation in midpoint and width, the two characteristics that differ from between compounds, and avoids much of the error due to scatter that would be introduced if T 50 were estimated by interpolation. T infl , the inflection point in the TD vaporization profile, 20 corresponds to the peak in the aerosol mass evaporation rate. T 50 tends to be slightly lower than T infl (by ∼1-2 • C) for pure compounds.
Vaporization profiles of mixtures reflect the volatility distribution and interactions among the components, as discussed below. Volatility distributions of mixtures have been studied previously in this laboratory using temperature-programmed thermal des-Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion and then the temperature is slowly increased as the mass spectrum of the evaporating material is monitored (Tobias and Ziemann, 1999). In TPTD, the signal intensity is proportional to the evaporation rate, and a desorption (TI signal vs. temperature) profile obtained using this technique is similar to the temperature derivative of a TD vaporization profile. Figure 4 shows T infl from TD vaporization profiles plotted against T des , the TPTD 5 peak desorption temperature, for several mono-and dicarboxylic acids and features in the vaporization profile for chamber-generated SOA from the reaction of pentadecane with OH (Lim and Ziemann, 2005). The TD T infl is uniformly higher than the TPTD T des by ∼16 • C, and after correcting for this temperature offset, the two techniques show very good agreement (the slope of the linear fit shown in Fig. 4 is 0.99±0.04). This al-10 lows TPTD desorption profiles to be used in the interpretation of ambient data obtained with the TD. A database of TD and TPTD vaporization profiles for various classes of chamber-generated SOA, including profiles for characteristic ions in many cases, is available online at http://cires.colorado.edu/jimenez-group/TDPBMSsd/ for use in the analysis of TD-AMS data. The similarity between the TPTD desorption profile and the 15 temperature derivative of a TD vaporization profile is illustrated in more detail below in Sect. 3.7.
3.2 log P 25 vs. T 50 calibration A plot of log P 25 vs. T −1 50 for the standard compounds used in this study is shown in  Table 1. The literature values used were restricted to studies in which the particles were generated by atomization of a solution, as they were for the particles used in the calibration, in order to avoid any bias due to the effect of residual solvent. The line is the linear least squares fit with errors in both T −1 50 and log P 25 taken 25 into account (York et al., 2004) and is given by the equation The standard deviation in log P 25 is ∼0.2, so the uncertainty in calculating P 25 for an unknown compound with similar particle size, shape and mass concentration from this curve should be roughly 0.2 orders of magnitude within the range covered by the model compounds, and increase somewhat with extrapolation. The model compounds consist of both solids and liquids, with a variety of functionalities (saturated dicarboxylic 5 acids, an unsaturated monocarboxylic acid, and a diester), showing that a reasonable fit can be obtained for a set of pure organic compounds with different physical and chemical properties. Since variations in temperature profiles can be expected for individual TDs, even those sharing the same design, the log P 25 vs. T −1 50 calibration may vary significantly from one TD to another. The set of standard compounds listed in Ta-10 ble 1 is well suited to the calibration of TDs to be used in vapor pressure measurements of atmospheric aerosol. Figure 6 shows measured values of T 50 for several aerosols, along with a shaded region encompassing the region 1 order of magnitude in P 25 above and below the calibration curve (Eq. 1). The aerosols represented in the figure are monocarboxylic 15 acids with particle diameters of 200 nm and mass concentrations of 150-200 µg m −3 , polydisperse oleic acid particles with a mass distribution peaking at ∼500 nm and mass concentration of ∼250 µg m −3 , and a laboratory-generated SOA from the reaction of pentadecane with OH (in the case of the SOA, T infl for features in the vaporization profile were used in place of T 50 ), as well as the standard compounds used in the calibration. 20 The literature values of P 25 used in the plot are listed in Tables 1 and 2, except those for the SOA features, which are based on a TPTD study of the same aerosol (Lim and Ziemann, 2005) and a calibration described below in Sect. 3.7. With the exception of the C 18 monocarboxylic acid, the literature values of P 25 for all the aerosols fall within 1 order of magnitude of the values predicted by the calibration. The generally low 25 values of T 50 for the monoacids may be due to differences in particle shape. Crystals of these compounds are often scaly, and it is possible that the particles they form by evaporation of the droplets from the atomizer are similarly thin and flat, and thus have a considerably greater surface area to volume ratio than the other particles. Interactive Discussion of variations in particle size and mass loading, as well as dilution with other compounds in a mixed particle, on T 50 , are addressed in more detail below. The spread in the literature values increases significantly with decreasing vapor pressure due to the difficulty in measuring very low vapor pressures, and values obtained by extrapolating to lower vapor pressures than those covered by the calibration (be-5 low ∼10 −6 Pa) are less reliable. Donahue et al. (2006) suggest that compounds with vapor pressures as low as 10 −8 Pa should be considered semivolatile. Estimating the vapor pressures of such compounds would entail extrapolating by about 2 orders of magnitude in P 25 , which could introduce a significant error. While it would be desirable to accurately estimate vapor pressures of ambient aerosols down to 10 −8 Pa, this will 10 only be possible when vapor pressures in this range are known with greater certainty.

Effects of particle size and mass concentration
Particle size and mass concentration affect both evaporation rates and equilibrium partitioning, and so are expected to influence the TD vaporization profiles obtained using this technique. Experiments and simulations were therefore performed to investigate 15 the dependence of T 50 on these quantities. T 50 was measured for oleic acid particles with diameters of 100, 200, 300, and 400 nm and several mass concentrations between 30 and 500 µg m −3 and simulated for the same particle diameters, and mass concentrations of 1-600 µg m −3 . Simulations of particle evaporation were performed using equations for the rate of change in particle diameter, d p , in the free-molecule (d p λ) and continuum (d p λ) regimes, where α, D v , P ∞ , P d , M, ρ are the evaporation coefficient, gas phase diffusion coefficient, partial pressure, equilibrium vapor pressure for a particle with diameter d , Interactive Discussion molecular weight, and condensed-phase density of the evaporating compound, T is the TD temperature in K, and R is the gas constant (Seinfeld and Pandis, 1998). The parameter values used in the simulations are given in Table 2. The parameters used for oleic acid in the simulation were altered somewhat from literature values and the effective residence time was reduced from 15 to 6.5 s for all simulations in this paper. 5 These changes are not unreasonable, since the model does not account for all the complexities of the system, and they yielded better fits to the data while not altering the major conclusions derived from the simulations. The integrated value of d p was calculated at intervals of 10 ms over the residence time of the aerosol in the heated region. T 50 was determined by varying the temperature and repeating the calculation 10 above until the fraction of mass remaining converged to 0.5 within a tolerance of 10 −6 . The effect of mass concentration was accounted for in the simulation by calculating P ∞ at each time step, using the mass of aerosol evaporated at that step, and assuming ideal behavior. The changes in the gas phase diffusion coefficient, the heat of vaporization, and the residence time in the heated region (due to thermal expansion) 15 with increasing temperature were accounted for. In these simulations the Kelvin effect was ignored, since even for the smallest oleic acid particle considered, one of 80 nm formed by evaporation of 50% of the mass from a 100 nm particle, the increase in the vapor pressure due to surface tension (assuming a value of 0.03 J m −2 from Tao and McMurry, 1989) is only ∼20%. 20 As shown in Fig. 7 for both the measurements and simulations, T 50 increases as either the particle size or the mass concentration increases. The effect of particle diameter on T 50 is apparent in the experimental data for mass concentrations up to at least 300-400 µg m −3 . The continuum model captures the trends in the data with respect to both particle diameter and mass concentration. For the 200 nm particles, both AMTD 1, 2008 Thermodenuderparticle beam mass spectrometer system Interactive Discussion evaporation rate by ∼20%. Not only are the Kelvin and non-continuum effects small, but they have compensating effects on evaporation rates. The good agreement between measurements and simulations provides support for the use of this model to explain and predict particle behavior in the TD. For example, some useful insights can be gained by considering the case where P ∞ is negligibly 5 small compared to P d . Integrating Eq. (3) explicitly with respect to t and solving for the case where d p /d p,0 =(1/2) 1/3 , the value of the diameter ratio when the initial mass has been reduced by 50%, gives the following equation

AMTD
where t r is the residence time in the TD. Without solving explicitly for T 50 , it is possible 10 to get some insight into its dependence on d p,0 by noting explicitly the temperature dependence of the particle vapor pressure, P d (T 50 ), as given by the Clausius-Clapeyron equation where ∆H vap is the heat of vaporization. Because P d (T 50 ) depends exponentially on 15 T 50 , the change in T 50 that occurs as the result of a change in d p,0 is determined primarily through the P d (T 50 ) term in Eq. (4) rather than T 50 in the denominator. Hence, if d p,0 is doubled, the factor of 4 increase in P d (T 50 )/T 50 introduced by the d 2 p,0 term is primarily compensated for by a proportionately much smaller increase in T 50 that is amplified through the P d (T 50 ) term. For example, at low aerosol mass concentrations 20 where P ∞ is very small and Eq. (4) is applicable, the ratio of T 50 values (in K) shown in Fig. 7 for 400 and 100 nm particles is only ∼1.05 while the square of the diameter ratio is 16. In addition, the increase in T 50 with increased aerosol mass concentration that is observed in Fig. 7 can be understood by noting that for a given initial and final particle diameter, more vapor is formed at a higher aerosol mass concentration. This 25 increases P ∞ , which decreases the evaporation rate according to Eq. (3), meaning that higher TD temperatures are required for particles to lose 50% of their mass. The log P 25 vs. T 50 calibration equation, as mentioned above, was calculated using data from particles with d p =200 nm and mass concentrations of 100-200 µg m −3 . The error incurred by using this calibration for particles with other diameters and mass concentrations can be estimated using the simulation results. As shown in Fig. 7, continuum model simulations indicate that T 50 values for particles with the same composition 5 and initial diameters and mass concentrations anywhere in the range from 100-400 nm and 1-600 µg m −3 will differ by less than ∼11 • C from those at 200 nm and 150 µg m −3 , which is roughly the average for the calibration particles. For this range of conditions, which captures those typically encountered in the atmosphere and in the laboratory, the maximum error incurred by calculating P 25 using the calibration (Eq. 1) and a measured value of T 50 that is uncorrected for particle size and mass concentration would therefore be about a factor of 9 in P 25 (this is based on an 11 • C difference at the low end of the T 50 range, where the change in log P 25 with T 50 is the greatest). The magnitude of the error for any complex aerosol will vary with particle composition, phase, morphology, and mixing state, factors that are generally unknown and are therefore 15 difficult or impossible to account for in simulations. Ambient organic particle mass concentrations are nearly always lower than the range used in the determination of the calibration curve given by Eq. (1), which raises the question of whether there would be advantages to determining a separate calibration curve at lower mass concentrations. However, the effect of particle size on the evaporation kinetics is most pronounced at 20 low mass concentrations, and at 10 µg m −3 , for example, might easily be more important than the correction for mass concentration. Therefore, separate calibration curves for different particle sizes or size distributions would be necessary. Since the maximum error incurred by using the current calibration is of a magnitude similar to the uncertainty in the calibration itself, such a refinement would offer little if any benefit for most The derivative of M T /M 0 for a mixture with respect to T −1 TD is a good proxy for a vapor pressure distribution, since the T TD at the median in the derivative of the signal for a particular compound is equal to T 50 , from which the vapor pressure can be calculated from the calibration curve. The distribution calculated in this way shows the relative 5 amount of condensed phase material vs. vapor pressure, and since the TI signal is approximately proportional to mass (Crable and Coggeshall, 1958), the intensity is proportional to the mass concentration. For a mixture of compounds, the vapor pressure distribution is a conceptually useful representation of the data that can be obtained with the TD-mass spectrometer. 10 To generate such a plot from a TD vaporization profile, the M T /M 0 curve is numerically differentiated with respect to T −1 TD , and the x-axis is then converted from T −1 TD to log P 25 using the log P 25 vs. T 50 calibration, i.e., Eq. (1), with T 50 replaced with T TD . Multiplying d (M T /M 0 )/d (T −1 TD ) by the Jacobian, which is simply the inverse of the slope in Eq. (1), yields the normalized log-scale mass vs. vapor pressure dis-15 tribution, M(log P 25 ). The intensity is, of course, convoluted with the shape of the TD vaporization profile for the individual components, and the vapor pressure of a component in a mixture is not generally equal to P 25 for the pure compound, but is affected by the mixing state. The effect of these approximations and others are discussed in detail in Sect. 3.5. Center-point differentiation (i.e., for data-point i , ) was found to be optimal for the experimental datasets in this study. Figure 8 shows ( Interactive Discussion perimental and simulated T 50 values for the same (literature or input) molecular properties. The TD vaporization profile was simulated using a continuum model as described above, and the parameters used in the calculation are shown in Table 2. The P 25 values and relative mass concentrations of the different compounds used in the simulation are shown as vertical lines in the log P 25 distribution. Some differences between the 5 input distribution and the distribution calculated from the TD vaporization profile are apparent, and will be discussed in detail below in the context of the binned log C * 25 distribution.

Volatility basis set analysis
A volatility distribution of the type used by Donahue et al. (2006), showing the con-10 centration and gas-particle partitioning of aerosol components as a function of C * 25 , the saturation concentration at 25 • C, and divided into bins based on log C * 25 (spaced, for example, by one order of magnitude in C * 25 ), can also be estimated from the TD vaporization profile. In contrast to the vapor pressure distribution described above, which shows only the concentration of condensed phase material, this volatility distri-15 bution also includes the concentration of gas phase material inferred using partitioning theory.
The procedure for converting the TD vaporization profile to the C * 25 distribution is illustrated in Fig. 8a and c. The fraction of a mixture (or single compound) vaporizing between any two temperatures is simply equal to the difference in M T /M 0 evaluated at 20 those temperatures; therefore the mass fraction f i of the particle-phase material in a mixture belonging in each log C * 25 bin can be calculated in this manner from the TD vaporization profile. First, it is necessary to determine the thermodenuder temperatures corresponding to the edges of each log C * 25 bin. For an ideal mixture, the saturation concentration of a compound in µg m −3 is given by where M and P • are the molecular weight in g mol −1 and partial vapor pressure in Pa AMTD Here, as in the calculation of M(log P 25 ), T 50 in Eq.
(1) has been replaced with T TD , and P 25 has been replaced with P • . In general, the identity of the compounds in the 5 mixture being analyzed is not known, and the basis set can be considered to represent a set of hypothetical compounds, with saturation concentrations spaced by a factor of 10 in C * 25 . M may be replaced with an estimated average molecular weight, or it may be treated as a function of C * , with the hypothetical compound in each log C * 25 bin having its own molecular weight M i . If a different molecular weight is used for each log C * 25 bin, f i must be adjusted using the Jacobian due to the non-linear dependence of log C * 25 on T TD . The calculation of f i is illustrated in Fig. 8a for the log C * 25 =1 bin, with the dashed lines indicating the values of T −1 TD , log C * 25 , and M T /M 0 at the edges of the bin. For the experimental datasets analyzed in this study, M T /M 0 at the temperatures corresponding to the boundaries of each log C * 25 bin were found by linear interpolation, 15 and a calculated or estimated average molecular weight was used.
Next, it is necessary to determine C p and C g , the particle-and gas-phase concentrations for the material in each log C * 25 bin. From partitioning theory (Donahue et al., 2006;Pankow, 1994a) 20 where C OA is the total concentration of particle-phase organic matter, which must be measured in a separate experiment or estimated. C p,i is equal to the fraction of the total C OA which belongs in bin i , i.e., Combining Eqs. (8) and (9) gives The values of C p,i are represented by the solid areas of the bars in Fig. 8c, and the values of C g,i are represented by the clear areas.
Gas-particle partitioning of aerosol prior to entering the TD will be determined by the ambient temperature; therefore if TD experiments are performed at an ambient temperature other than 25 • C Eqs. (9) and (10) will give the particle and gas phase 5 concentrations for compound i at that ambient temperature, and C * i in Eq. (10) must be the saturation concentration for compound i at ambient temperature for the results to be valid. Therefore, the procedure is to first calculate the distribution at ambient temperature, then calculate the partitioning for the resulting total mass concentrations in each bin at 25 • C. To simplify the eventual conversion from the distribution at ambient 10 temperature to one at 25 • C, it is simplest to calculate f i for bins corresponding to the C * i 25 basis set, that is, to keep the same set of hypothetical compounds. The log P 25 vs. T −1 50 calibration will still be valid, and f i and C p,i can be calculated as described above.
The C * i values at ambient temperature that correspond to the C * i 25 basis set values can be calculated using the Clausius-Clapeyron equation (Eq. 5) and the fact that C * 15 is proportional to vapor pressure (Eq. 6), which combine to give where T amb is the ambient temperature in K. In Fig. 8a this would be equivalent to changing the log C * 25 axis to a log C * T,amb axis, but keeping the dashed lines defining the bin edges fixed. Donahue et al. (2006) suggest using values of ∆H vap that decrease 20 with increasing C * , with ∆H vap =100 kJ/mol for C * =1 µg m −3 at 300 K, and an increment of −5.8 kJ mol −1 for each successive log C * bin, when the bins are separated by a factor of 10 in C * . Once C p,i and C g,i for each log C * T,amb bin have been calculated using Eqs. (9) and (10), the total concentration of organic material for each log C Equations (12) and (13) can be iteratively solved to find the volatility basis set distribution at 25 • C (Donahue et al., 2006). Volatility information from the TD extends up to the C * corresponding to the ambient temperature. 5 Several factors that influence the measured volatility distributions can be seen by comparing the input distribution ("simulation input") and the distribution calculated from the simulated TD vaporization profile ("simulation output") in Fig. 8c. The width of the TD vaporization profile, even for a pure compound, will broaden the measured distribution. For typical TD vaporization profiles of pure standards, M T /M 0 12 • C above and 10 below T 50 is ∼0.9 and 0.1, respectively. The broadening in the calculated C * distribution increases with decreasing T 50 . For T 50 =40 • C, ∼10% of the mass will be calculated to be at a C * 1 order of magnitude higher, and ∼10% 1 order of magnitude lower, than the true C * . For T 50 =170 • C, the difference is reduced to about 0.5 orders of magnitude. The output distribution in Fig. 8c shows significant intensity in the 10 2 µg m −3 15 bin, where there is none for the input distribution, due to this effect. In addition, there are factors which bias the TD vaporization profile of each component in a mixture, that is, the plot of the mass of that component in the particle phase divided by its initial mass vs. T TD , relative to the TD vaporization profile of particles of the pure compound at the same initial particle size and number concentration. Since 20 the total TD vaporization profile for a mixture calculated from the TI signal is essentially the mass fraction weighted average of the component profiles, this is an appropriate comparison. Differences in the partial vapor pressure are one such factor. Initially, if we assume ideal behavior, the partial vapor pressure of a component is equal to its vapor pressure in a pure particle multiplied by its initial mole fraction in the mixture.

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion rate are reduced by a similar factor. As material evaporates from the particle, however, the mole fraction, and therefore the partial vapor pressure, will be reduced for more volatile components and increased for less volatile components, relative to that in the mixed particle initially. This causes more volatile components to tail toward lower volatility, and less volatile components to be shifted toward higher volatility, causing a 5 bias toward the center of the distribution and a shift toward higher volatility of the low volatility cutoff. At the same time, the particle size at a given point in the TD vaporization profile for a specific component is affected as the particle composition is changed by evaporation. For high volatility components, the evaporating particle will be larger for a mixture than for a pure particle due to the remaining low volatility material, and 10 for low volatility components, it will be smaller since the particle has already shrunk due to the removal of higher volatility species by the time the low volatility species are evaporating significantly. This increases or decreases, respectively, the surface area available for evaporation for high and low volatility components (since we are comparing vaporization profiles for the same number density of particles), causing a bias that 15 is opposite to, but less than that of the partial vapor pressure (the actual effect of particle surface area on the rate of mass lost from the particle is particle size-dependent, but it is less important than the effect of the changing partial vapor pressure in either the continuum model or the free molecule model). In Fig. 8c, the combined effect of these factors is less obvious at the high volatility end of the distribution, but can be seen 20 clearly at the low volatility end, where the simulation output shows much less mass in the 10 −2 µg m −3 bin than the input distribution does.
Of the factors discussed above, the broadening due to the TD vaporization profile width is probably the most significant. It will tend to be most obvious at the high vapor pressure end of the distribution, where it is greater and there are no significant op- 25 posing effects, and may lead to large errors in the total mass assigned to high C * bins, since the C g /C p ratio is highest there. While there is no fool-proof way to correct for this, intensity in bins at the high C * end of the distribution should be treated with caution, especially when the intensity in the bins immediately to lower C * is much greater. structures and a large fraction of liquid oleic acid, was chosen in order to increase the likelihood of the particles being a single liquid phase. The particles were 200 nm in diameter and the total mass concentration was 100-150 µg m −3 , similar to the conditions used to generate the calibration curve. In one experiment, mass fragments characteristic of each of the acids were monitored in SIM mode, and in another, full spectral 10 scans were recorded and the TI signal computed. The vapor pressure distributions calculated from the characteristic mass fragments, the TI signal, and the mass fraction weighted average of the characteristic fragment signals are shown in Fig. 9a. The individual fragment distributions are scaled by a factor of 1/2 for clarity. The top axis shows the log P 25 scale calculated using Eq. (1), 15 and the vertical lines indicate the log P 25 values for the pure individual compounds from the literature, which are listed in Table 2. The C 15 and C 16 monoacid profiles exhibit the expected ordering, with the C 16 compound evaporating at a slightly higher temperature than the C 15 , and the peaks in their signals agree reasonably well with the literature P 25 values. The SIM curves are wider than those typically observed for pure 20 compounds, with the curve for the C 15 monoacid tailing toward higher temperature and the other curves broadened in both directions. Nonetheless, on the low temperature side of the curves the TI or sum of SIM signals provide good approximations of the vapor pressure distribution. The curves for the less volatile components do not follow the behavior expected from their vapor pressures; rather, all three peak at essentially 25 the same temperature, corresponding to a P 25 value close to that of oleic acid. Similar volatility behavior has been seen previously in monoacid and diacid mixtures containing oleic acid (Chattopadhyay, 2004), and suggests non-ideal behavior of the mixture. The

AMTD Introduction
Conclusions References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion similarity of the TD vaporization profiles of oleic acid and the C 18 and C 20 monoacids suggests that they may form a separate phase, excluding the other two components, with oleic acid acting as a matrix which determines the volatility behavior of the phase. The TD vaporization profiles of the C 18 and C 20 monoacids in this mixture reflect their effective vapor pressures in the mixture in the temperature range in which they evapo-5 rate significantly. The effective vapor pressure of a component in a mixture is of interest in itself, since it determines the volatility behavior of the component as long as the mixture in which it is present is fairly constant. Between 25 • C and this temperature range, the organization of the mixture among condensed phases may change, so it is not clear whether the effective P 25 values for oleic acid and the C 18 and C 20 monoacids in 10 this mixture can be calculated from the calibration (Eq. 1). The much more complex mixtures typically found in ambient aerosol are less likely to show such behavior, since they are more likely to consist of a complex mixture in which no single compound is present in such a high concentration that it acts as a matrix. In Fig. 9b, the values of C p (solid area) and C g (empty area) calculated from the 15 weighted sum of fragment signals, binned by order of magnitude in C * , are shown, along with the true distribution calculated from the mass fractions of components in the mixture and literature values of P 25 , and the distribution recovered by simulating the TI signal for the mixture with a continuum model, using the true distribution as input. As in Fig. 8, a separate calibration was used to calculate P 25 and C * for the simulation output, 20 so that differences between the distributions calculated from the experimental vaporization profile and the simulation output more closely reflect differences between the real volatility behavior of the mixture and simulated ideal behavior, rather than biases in the simulation. C * in this plot is calculated from P 25 using an averaged molecular weight and assuming ideal behavior. The experimental distribution shows significant intensity 25 in the 10 2 µg m −3 bin, where there is none for the true distribution. This is consistent with the behavior seen for the hypothetical distribution shown in Fig. 8, and the fact that the simulation output shows the same behavior, although to a somewhat lesser extent, supports the conclusion that this is due to the finite width of the vaporization profile for the C 15 acid. The low volatility side of the distribution for both the experimental distribution and the simulation output is biased toward higher volatility -neither shows intensity in the 10 −2 µg m −2 bin, where there is significant intensity in the true distribution, and from the intensities in the 10 −1 µg m −3 bin, it is apparent that the bias is even greater for the experimental distribution than for the simulation output. This is 5 the largest difference between the experimental and simulated distributions, and probably reflects the non-ideal behavior described above, in which the three lowest-volatility components vaporize at essentially the same temperature. Considering uncertainties in the literature values for P 25 (values shown in Table 1 for individual components in this mixture vary by a factor of ∼2-5), the agreement between the experimental distribution 10 and the simulation output is otherwise quite good.

Secondary organic aerosol
Chamber-generated SOA, though less complex than ambient aerosol, is still much more complex than the monoacid mixture discussed above. The volatility of SOA formed from the reaction of pentadecane with OH radicals in the presence of NO x 15 has been studied previously in this laboratory (Lim and Ziemann, 2005), using TPTD. Two fairly well-defined peaks and a shoulder were seen in the desorption profile, which makes this a particularly good system for evaluating the TD method. A calibration of log P 25 vs. T −1 des for the TPTD technique was determined using a series of saturated mono-and dicarboxylic acids, with P 25 determined from the single-20 compound desorption profiles (Chattopadhyay and Ziemann, 2005), and the equation of the least squares fit to all the data points was Note that the slope of this equation is similar to that in the TD calibration curve (Eq. 1). The TPTD desorption profile and the temperature derivative of the TD vaporization 25 profile for SOA formed from the pentadecane + OH reaction are shown in the top and bottom panels of Fig. 10, respectively. The temperature axes are offset by 16 • C for AMTD 1,2008 Thermodenuderparticle beam mass spectrometer system values of P 25 measured by the two techniques are within a factor of ∼3 for each peak, which is well within the estimated uncertainty in calculating P 25 . There are significant differences in the relative intensities of the various peaks that may reflect differences between the techniques or real differences in the composition of the aerosol, which may vary somewhat between experiments. Overall, the consistency between the two 10 methods is quite good. The log P 25 distribution and binned C * distribution calculated from the TD vaporization profile are shown in Fig. 11a and b. The two major features in the log P 25 distribution, centered at log P 25 =−4 and −8 (log C * ∼ =1 and −3), are still visible in the log C * distribution after binning. The small intensity in the 10 2 µg m −3 bin is probably due to 15 the broadening of the signal from material in the 10 1 µg m −3 bin, in which the intensity is much higher. The intensity in the 10 1 µg m −3 bin, however, is probably a good indication of the true amount of material in that bin. The TD vaporization profile for this SOA sample was measured at a particle mass concentration of ∼150 µg m −3 , which is much higher than typical ambient SOA concen-20 trations. The volatility distribution predicted for this SOA sample after 10-fold dilution, found by solving Eqs. (12) and (13) iteratively for C OA and C p,i , is shown in Fig. 11c. The particle mass concentration, C OA , is reduced from 150 µg m −3 to 13.4 µg m −3 (a slightly greater than 10-fold decrease, due to the greater fraction of mass in the gas phase at higher dilution), and the increase in the fraction of material in the gas phase for C * >10 −1 µg m −3 is evident.

Mass spectral analysis
The composition of aerosol as a function of volatility is of considerable interest in learning about how the volatility distribution changes with photochemical aging, and it may also enhance the separation of OA sources/components for component analysis methods that identify sources and components by exploiting mass spectral differences 5 (Zhang et al., 2005;Ulbrich et al., 2008). Differences in the mass spectrum as the composition of the vaporized fraction changes may also yield information on the composition of the different volatility fractions (Huffman et al., 2008a 1 ). In the case of the SOA generated from the pentadecane + OH reaction, the presence of well-defined peaks in the log P 25 distribution in Fig. 11a suggests the possibility of comparing the 10 mass spectra obtained at the temperatures corresponding to these peaks. Figure 13 shows mass spectra of the vaporized fraction (that is, the difference between the spectra measured when the aerosol is sampled at the exits of the bypass tube and the TD, respectively) at T TD =45 and 100 • C, corresponding to log P 25 =−3.9 and −7.7, respectively, the positions of the two most prominent peaks in the vapor pressure distribution.

15
Peaks at m/z 225, 239, 241, and 286, which are absent at 45 • C, can be seen at 100 • C. This is consistent with the mass spectra obtained at the corresponding peaks in the TPTD experiment (Lim and Ziemann, 2005), and shows that it is possible to obtain information on the chemical composition of aerosol as a function of volatility using this technique.

Conclusions
This paper describes the development and evaluation of a technique that couples a thermodenuder with a particle beam mass spectrometer to determine the vapor pressures of organic aerosol components. An important feature of this technique is its simplicity, which allows the vapor pressure distribution for a complex mixture such as Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion a single calibration curve. The empirical approach avoids complex modeling and the need to make assumptions about numerous unknown properties of the aerosol and physical parameters of the system. While ignoring these complex problems does not make them go away, the range of uncertainties that are likely to be encountered in the application of this method can be explored by studying realistic systems. This has 5 been attempted here by using simulations and by analyzing a simple, five-component mixture and a more complex chamber-generated SOA. The results suggest that for the range of particle sizes and mass concentrations typical for the atmosphere and laboratory studies, vapor pressures of aerosol components can probably be estimated to within about one order of magnitude, which is accurate enough to be of consider-10 able use in aerosol volatility studies, and is a vast improvement over the estimates currently used in atmospheric models (Huffman et al., 2008a 1 ). Volatility distributions using the volatility basis set approach of Donahue et al. (2006) can be estimated easily from the TD vaporization data, implying that the TD-AMS will be of use in modeling based on this type of volatility analysis. From the experiments on simple and complex 15 (SOA) mixtures, it is also evident that some separation of compounds by volatility can be achieved, and that it is possible to obtain information on aerosol composition as a function of volatility. This may be of considerable interest for the development of methods for deconvoluting AMS spectra of different organic aerosol classes (Zhang et al., 2005), which are important for advancing the analysis and understanding of organic 20 aerosols, and for studying the evolution of aerosol volatility with photochemical aging.
carboxylic Acids and C5-Dicarboxylic and C6-Dicarboxylic Acids, Environ. Sci. Tech., 23, 1519-1523, 1989 A.: Atmospheric Photo-Dissociation Lifetimes for Nitromethane, Methyl Nitrite, and Methyl Nitrate, Int. J. Chem. Kinetics, 12, 231-240, 1980. 25 Tobias, H. J. and Ziemann, P. J.: Compound identification in organic aerosols using temperature-programmed thermal desorption particle beam mass spectrometry, Anal. Chem., 71, 3428-3435, 1999. Tobias, H. J., Kooiman, P. M., Docherty, K. S., and Ziemann, P. J.: Real-time chemical analysis of organic aerosols using a thermal desorption particle beam mass spectrometer, Aerosol AMTD      (1). The vertical bars in (b) indicate the vapor pressures and mass fractions of the compounds used as input for the simulation. The mass fractions of the particle mass concentration belonging to each order of magnitude C * 25 bin, necessary for the volatility basis set analysis (c) are calculated by taking the difference between M T /M 0 at the edges of the bin; the dashed lines in (a) indicate those values for the C * 25 =10 1 bin. Filled and empty areas of the bars indicate particle phase and gas phase material, respectively. The distribution shown by solid bars in (c) was calculated from the curve in (a) by this procedure, and the distribution shown by the patterned bars in (c) was used as input for the simulation. Introduction  Table 2 for each compound. (b) Volatility distribution for the mixture of C 15 , C 16 , C 18 , and C 20 monoacids and oleic acid showing calculated gas-particle partitioning. Filled and empty areas of the bars indicate particle phase and gas phase material, respectively. The experimental distribution was calculated from the mass fraction weighted average of the SIM profiles. The true distribution was calculated from the mass fractions of the components in the mixture and the literature values of P 25  Volatility basis set distribution for laboratory-generated SOA formed from the reaction of pentadecane with OH radicals. (c) Calculated volatility distribution for the same aerosol after 10-fold dilution. Filled and empty areas of the bars in (b) and (c) indicate particle phase and gas phase material, respectively. The particle mass concentration was ∼150 µg m −3 .
AMTD 1,2008 Thermodenuderparticle beam mass spectrometer system Mass spectra of material volatilized at 45 • C and 100 • C from SOA formed from the reaction of pentadecane with OH radicals. The spectra were calculated by subtracting the mass spectrum of aerosol sampled after passing through the TD from that sampled after passing through the TD bypass tube.